Schroedinger

The difference between 65nm and 45nm is not that 45nm deals with heat better. It just doesn't create heat as copiously. Fans, water, etc. just diffuse heat; they don't reduce heat below ambient. 45nm doesn't create as much heat above ambient anyway, so diffusion based cooling with the ambient temperature isn't very effective. More to the point, at ambient, heat is not the limiting factor on an overclock.

Sub-Ambient cooling increases the efficiency of the transistors. Just as 50 degrees C is better than 100, 0 or -100 is better than 50 degrees C. Sub Ambient cooling still Significantly changes the temperature at which a 45nm processor operates which is of course the relevant efficiency determining factor.

Heat is every bit as important to a 45nm processor as to a 65nm processor. The difference is not the effect of heat, but the production of heat. On good air cooling (IE - diffusive cooling to ambient temps) 45nm processors will run into voltage problems before they run into heat problems, at least potentially.

On Sub Ambient cooling, the same thing is true, but temperature at the core is very very low, so there are switchign efficiency gains that translate into higher OCs.

I do NOT believe this should justify higher voltages for sub ambient cooling, although I'll defer to more knowledgeable parties on that score.

Wankerfx

That is true Mr. Shroedinger, the cat killer .

But it has been proven on several accounts that electromigration is more to fear than temperatures. I could see the cold preserving the circuits, but it won't preserve them enough to notice a big difference. In the end, air cooling has a better price : performance ratio in comparison to any other source of cooling. (For 45nm only)

__________________

xdaseinx

Yea but when a processor runs faster it needs that added voltage. The transistor gate is either on or off based on a reading of voltage thru the gate and the vcore setting, meaning when the gate is checked if it has a certain amount of voltage it is read an on. When you overclock, the voltage has less time to drop in the gate, and often a gate that is off is read as on. That's why you pump more vcore. So there's a higher treshhold for an on and less chance for a false on.

Edit:
What I'm saying is, you'll benefit from sub zero temps as you mention (though, I'm still not too clear on why), but not by much because you still need to pump more volts if you want performance at high clocks. But can't due to electromigration which as mentioned is the problem with the 45's and not temperature.

__________________

Schroedinger

I didn't mean to contradict that. LN cooling is pretty silly except for "because I can" benchmarking. For supercomputer tasks you can just link more processors, and I can't think of two thread tasks that require much more than 4.0ghz right now...

I agree about air, and it's great that Intel has now made a chip that deals with heat problems simply by process efficiency. I was responding to the previous poster's specific question about why sub ambient cooling was still relevant.

As I said at the end, my guess is that on any chip, sub ambient cooling doesn't actually allow higher voltages from an electromigration standpoint. It just makes the chip function better, period, and obviates the heat problems which are a secondary effect of overvolting a part.

Edit: I did further research. As you approach -273C, superconductive properties reduce electromigration as well as increasing switching efficiency. As a result, LN should decrease the damage caused by overvoltage, although it should not in theory prevent the damage entirely. I left the above as is, I'm not trying to say I know everything!

In response to the above posters edit: Sub Ambient cooling, therefore, allows you to BOTH do less damage via electromigration with the same voltage AND get better performance (higher clock speed, more stability) at any given voltage.

Schroedinger

[quote=Wankerfx;3381496]That is true Mr. Shroedinger, the cat killer .

QUOTE]

You don't know that. There's only a 50% probability that my cat is dead. In a sense, it's both alive and dead right now. It's like a zombie cat.

(Sorry, this is a rather esoteric reference).

xlink

this was the same thing when 65nm cedar mills came out.

1.6V on a cedar mill was like 1.9 on a prescott.

it's only logical to think that the same would apply here. I'll say this though, anandtech, noobville that they are was right when they said that 45nm changed the game. It's about performance/watt now... or should I say performance/lifespan, because the two are seemingly directly proportionate at this point and heat has relatively little to do with it.

I've had this long standing practice. Stock volts over clock from day 1. When I feel i want a new chip, buy a $10 backup off ebay and then on your main chip go as high as Intel/AMD rates the chip and run at that for a while under the assumption that it will die at any time.

lordikon

I'm confused how this is necessarily a bad thing?

Chips have always fried at high percentage voltage increases, the 45nm chip is no different.

The only difference is now it is much easier to cool.

All this means is people need to know the max voltages of new chips and adhere to them, and they can now save a crapload of money by using air cooling systems.

We may find new air cooling setups that have quieter/slower fans once again, that would be nice, running 5.5Ghz on a 32nm 16-core chip with a heatsink that doesn't weigh 5lbs, and has a fan running at 1500rpm.

__________________
3DMark06: 17933, Vista -- Super-pi 1M: 12.471s -- Crysis GPU Bench Vista, 1280x1024 No AA: Avg FPS, 57.45

opt33

Sounds sweet to me, and add a gpu that only needs a fan at 1300 rpms, and I would be happy to go back to air myself.

After being used to near dead silence, when i rma'ed my ultra I had to use an old 7950 with a fan for two weeks, drove me nuts.

RussianGrimmReaper

A stock E8400 with an IFX-14+IFX-10, you don't need anything more than an 800rpm fan

__________________

opt33

Try 1.4 volts at 4.2 ghz in ambients of 28C, and then let me know how that works.

And what does that have to do with my remark of cooling my OC gpu silently? Reading comprehension on some of these forums is scary.